`SelectSwapQROM` revamp and upgrades #986

tanujkhattar · 2024-05-22T17:08:29Z

Fixes #108
Partially addresses #368
Fixes #78

We still need an adjoint bloq for QROMs to make the complexity for clean ancilla version match that of the paper, and I haven't implemented the optimization to save one target register from https://quantum-journal.org/papers/q-2019-12-02-208/pdf/

This PR introduces major revamp to SelectSwapQROM and some upgrades to QROM. I'll add more detailed description soon

@fdmalone

…ire_symbol_to_cirq_diagram_info

…oam_updates

fdmalone · 2024-05-22T22:45:58Z

qualtran/bloqs/data_loading/qrom_abc.py

+
+@cirq.value_equality(distinct_child_types=True)
+@attrs.frozen
+class QROMABC(abc.ABC):


QROMInterface or something else, I'm not mad on ABC

BaseQROM or _BaseQROM or QROMBase

fdmalone · 2024-05-22T22:46:23Z

qualtran/bloqs/data_loading/qrom_abc.py

+@cirq.value_equality(distinct_child_types=True)
+@attrs.frozen
+class QROMABC(abc.ABC):
+    r"""Bloq to load `data[l]` in the target register when the selection stores an index `l`.


See #988 for including the abstract base class docs in a notebook somewhere.

fdmalone · 2024-05-22T22:47:35Z

qualtran/bloqs/data_loading/qrom_abc.py

+            |d_L[s_1, s_2, \dots, s_k]\rangle
+    $$
+
+    There two high level parameters that control the behavior of a QROM are -


nit: There -> The

fdmalone · 2024-05-22T22:47:57Z

qualtran/bloqs/data_loading/qrom_abc.py

+
+    There two high level parameters that control the behavior of a QROM are -
+
+    1. Shape of the classical dataset to be loaded ($\text{data.shape} = (S_1, S_2, ..., S_K)$).


nit: The shape, The number

nit: datasets

fdmalone · 2024-05-22T22:49:30Z

qualtran/bloqs/data_loading/qrom_abc.py

+       to build the QROM.
+
+    ### Shape of the classical dataset to be loaded.
+    QROM bloq supports loading multidimensional classical datasets. In order to load a data


Not The or A

Can you clarify?

I'm not sure anymore lol. I think it's was meant to be nit: The (or A) QROM bloq supports...

fdmalone · 2024-05-22T22:49:56Z

qualtran/bloqs/data_loading/qrom_abc.py

+    registers with bitsizes $(p, q, r, s)$ where
+    $p,q,r,s=\log_2{P}, \log_2{Q}, \log_2{R}, \log_2{S}$.
+
+    In general, to load K dimensional data, we use K named selection registers `(selection0,


nit: $K$ etc

fdmalone · 2024-05-22T22:50:32Z

qualtran/bloqs/data_loading/qrom_abc.py

+
+    If you have multiple classical datasets `(data_1, data_2, data_3, ..., data_L)` to be loaded
+    and each of them has the same shape `(data_1.shape == data_2.shape == ... == data_L.shape)`
+    and different target bitsizes `(b_1, b_2, ..., b_L)`, then one construct a single classical


nit: constructs

fdmalone · 2024-05-22T22:51:27Z

qualtran/bloqs/data_loading/qrom_abc.py

+    - `data.shape == data_1.shape == data_2.shape == ... == data_L` and
+    - `data[idx] = f'{data_1[idx]!0{b_1}b}' + f'{data_2[idx]!0{b_2}b}' + ... + f'{data_L[idx]!0{b_L}b}'`
+
+    Thus, the target bitsize of the merged dataset is $b = b_1 + b_2 + \dots + b_L$ and clifford


nit: the clifford cost

fdmalone · 2024-05-22T22:51:36Z

qualtran/bloqs/data_loading/qrom_abc.py

+    - `data[idx] = f'{data_1[idx]!0{b_1}b}' + f'{data_2[idx]!0{b_2}b}' + ... + f'{data_L[idx]!0{b_L}b}'`
+
+    Thus, the target bitsize of the merged dataset is $b = b_1 + b_2 + \dots + b_L$ and clifford
+    cost of loading merged dataset scales as


nit: the merged

fdmalone · 2024-05-23T00:14:00Z

qualtran/bloqs/data_loading/select_swap_qrom.py

+
+
+@bloq_example
+def _qroam_symb_dirty() -> SelectSwapQROM:


I think for the purposes of exposition it would be helpful to add a 1D example with the symbols matching those from the screenshotted costs.

fdmalone · 2024-05-23T00:14:25Z

qualtran/bloqs/data_loading/select_swap_qrom_test.py

        )
        for didx, data in enumerate([[[1, 2, 3, 4, 5]], [[1, 2, 3], [3, 2, 1]]])
-        for block_size in [None, 1, 2, 3]
+        for block_size in [None, 0, 1]


Did this get slower?

no, the different is that we now specify log_block_size instead of block_size; where block_size = 2**log_block_size.

fdmalone · 2024-05-23T00:15:40Z

qualtran/bloqs/data_loading/select_swap_qrom.ipynb

+    "\n",
+    "The `SelectSwapQROM` is a hybrid of the following two existing primitives:\n",
+    "\n",
+    "    * Unary Iteration based `QROM` requires O(N) T-gates to load `N` data\n",


This doesn't format well in the notebook.

fdmalone

Awesome! Should we just rename it QROAM at this point?

fdmalone · 2024-05-23T00:18:38Z

qualtran/bloqs/data_loading/qrom.py

@@ -54,7 +52,7 @@ def _to_tuple(x: Iterable[NDArray]) -> Sequence[NDArray]:

 @cirq.value_equality()
 @attrs.frozen
-class QROM(UnaryIterationGate):
+class QROM(QROMABC, UnaryIterationGate):


Would it make sense to ever just have a single QROM bloq where you'd get QROAM if block_size > 1, or would it make things more complicated.

mpharrigan · 2024-05-23T19:59:44Z

qualtran/bloqs/data_loading/qrom_abc.py

+from qualtran.simulation.classical_sim import ClassicalValT
+from qualtran.symbolics import bit_length, is_symbolic, shape, Shaped, SymbolicInt
+
+QROMT = TypeVar('QROMT', bound='QROMABC')


mpharrigan · 2024-05-23T20:07:28Z

related #699

…oam_updates

tanujkhattar · 2024-05-28T19:45:13Z

I've rewritten the docs on qrom base and derived classes and used the docs from base class in the notebooks for derived classes. I think we can still continue to improve the docs, but I'll go ahead and merge this for now.

fdmalone · 2024-07-30T18:04:34Z

@tanujkhattar I remember discussing this at the time, but what's the origin in the factor of 2 difference between the ceiling term when comparing Theorem 1 and 2, I think this is not related to the inverse right?

tanujkhattar · 2024-07-30T18:20:08Z

Cost for the dirty ancilla case (Theorem 1) matches the Appendix A formula (2 QROMs + 4 SwapWithZero's)

Cost for the clean ancilla case is different because we need

QROM + QROM.adjoint()
SwapWithZero + SwapWithZero.adjoint()

Our implementation assumes that the clean ancilla are freed up as part of the decomposition. In their implementation, they only count the cost of computing the table lookup; and measure the clean ancilla in X basis and keep the measurement results around as part of a deferred uncomputation strategy. Once the QROM needs to be uncomputed, these measurement results are used along with measurement based uncomputation of table lookup described in Appendix C.

Therefore, they essentially don't incur any cost for QROM.adjoint() and SwapWithZero.adjoint() -- therefore we have a 2x overhead in the total cost compared to their cost.

fdmalone · 2024-07-30T18:29:21Z

Thank you!

tanujkhattar added 6 commits May 13, 2024 17:49

SwapWithZero learns how to swap multidimensional selection index.

25cff73

Clear output

bbd9dca

Fix mypy and use wire_symbol instead of cirq diagram. Fix a bug in _w…

6db1485

…ire_symbol_to_cirq_diagram_info

Merge branch 'main' of https://github.com/quantumlib/Qualtran into qr…

3967587

…oam_updates

SelectSwapQROM revamp and upgrades

d280c16

Merge branch 'main' of https://github.com/quantumlib/Qualtran into qr…

613f489

…oam_updates

tanujkhattar requested a review from fdmalone May 22, 2024 17:08

tanujkhattar assigned fdmalone May 22, 2024

Fix pylint and typos

b10416b

fdmalone reviewed May 22, 2024

View reviewed changes

fdmalone reviewed May 23, 2024

View reviewed changes

fdmalone requested changes May 23, 2024

View reviewed changes

fdmalone reviewed May 23, 2024

View reviewed changes

mpharrigan reviewed May 23, 2024

View reviewed changes

tanujkhattar added 3 commits May 25, 2024 15:47

Merge branch 'main' of https://github.com/quantumlib/Qualtran into qr…

36503e2

…oam_updates

Fix mypy errors and address comments

3fcc199

Add type ignores, fix notebook and nits

9a2bad2

tanujkhattar mentioned this pull request May 28, 2024

Mypy issues in QROM / SelectSwapQROM #995

Open

2 tasks

fdmalone approved these changes May 28, 2024

View reviewed changes

tanujkhattar added 2 commits May 28, 2024 15:43

Merge branch 'main' of https://github.com/quantumlib/Qualtran into qr…

41de200

…oam_updates

Update docstrings and regenerate ipynb files

71fe8e8

tanujkhattar enabled auto-merge (squash) May 28, 2024 19:45

tanujkhattar merged commit cdf09f2 into quantumlib:main May 28, 2024
7 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

`SelectSwapQROM` revamp and upgrades #986

`SelectSwapQROM` revamp and upgrades #986

tanujkhattar commented May 22, 2024 •

edited

Loading

fdmalone May 22, 2024

mpharrigan May 23, 2024

fdmalone May 22, 2024

tanujkhattar May 28, 2024

fdmalone May 22, 2024

fdmalone May 22, 2024

fdmalone May 22, 2024

fdmalone May 22, 2024

tanujkhattar May 25, 2024

fdmalone May 28, 2024

fdmalone May 22, 2024

fdmalone May 22, 2024

fdmalone May 22, 2024

fdmalone May 22, 2024

fdmalone May 23, 2024

fdmalone May 23, 2024

tanujkhattar May 28, 2024

fdmalone May 23, 2024

tanujkhattar May 28, 2024

fdmalone left a comment

fdmalone May 23, 2024 •

edited

Loading

mpharrigan May 23, 2024

mpharrigan commented May 23, 2024

tanujkhattar commented May 28, 2024

fdmalone commented Jul 30, 2024

tanujkhattar commented Jul 30, 2024

fdmalone commented Jul 30, 2024


		There two high level parameters that control the behavior of a QROM are -

		1. Shape of the classical dataset to be loaded ($\text{data.shape} = (S_1, S_2, ..., S_K)$).

SelectSwapQROM revamp and upgrades #986

SelectSwapQROM revamp and upgrades #986

Conversation

tanujkhattar commented May 22, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

fdmalone left a comment

Choose a reason for hiding this comment

fdmalone May 23, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mpharrigan commented May 23, 2024

tanujkhattar commented May 28, 2024

fdmalone commented Jul 30, 2024

tanujkhattar commented Jul 30, 2024

fdmalone commented Jul 30, 2024

`SelectSwapQROM` revamp and upgrades #986

`SelectSwapQROM` revamp and upgrades #986

tanujkhattar commented May 22, 2024 •

edited

Loading

fdmalone May 23, 2024 •

edited

Loading